Referential Density in Typological Perspective

Referential Density in Typological Perspective Balthasar Bickel www.uni-leipzig.de/~bickel

An anthropological observation Belhare Pear Story (Sino-Tibetan; Nepal; Bickel 2003) pʌila .. aː .... ambibu

phighe

first

mango

otutuiʔ=na

jhola-e ukthe

quite.big=a

picked.from.above

illam

leŋse

basket-in put

il-lam

sassaba

from.there from.there by.pulling

riksa,

rikshaw

and.then

bag-in took.down

inetnahuŋgo dhaki-e and.then

kinahuŋgo

leŋse ʌni ... lead

and.then

eː saikil-lamma, saikil-lamma bicycle-on

bicycle-on

tahe

came

kinahuŋgo... and.then

2

An anthropological observation Maithili Pear Story (Indo-European; Nepal; Bickel 2003) ekṭā ām-ke

gāch

rahai. ā... a... a....

one mango-of tree

ām

me ek,

egoṭā chaurā

mango in

one

u

toir-ke

ām

is

one

boy

ām

mango

and.then

laḍkā

chaurā

one

boy

sāikal par

boy:HON bike

on

to.keep

jāi

chelai.

going was

elai,

came

caḍhne, to.ride

rahai

plucking is

ṭokari me rakhne

s/he mango having.plucked basket in

omaharse egoṭā

torait

ā...

3

RD measurement RD =

N (overt argument NPs) N (available argument positions)

RD measurements on Pear Stories (Chafe 1980), 10 speakers per language 7 languages (as of now, more in the works) 4

RD measurement (cont’d) •

Exclude metapragmatic comments and reported speech verbs.

•

Count available arguments per verb form, count complex predicates and serial verbs as one verb.

Belhare seu pheri leŋ-si

apple again direct-SUPINE

un

khaʔ-yu.

3sNOM

[3sS-]go-NPST

S/he/someone went again to put apples [into the basket].

Maithili okrā lālac 3hDAT

greed

āb-ait

come-IPFV.PART

chai.

is

S/he/they is/are getting greedy.

5


Exclude metapragmatic comments and reported speech verbs

•


•

Count existentials and equationals as licensing one argument position.

6



•


•


•

Count as overt NPs any NP (pronoun, demonstrative, noun, proper name etc.), including complex NPs with relative clauses.

7



•


•


•

Count as overt NPs any NP (pronoun, demonstrative, noun, proper name etc.), including complex NPs with relative clauses.

•

Delimite arguments from adjuncts through semantics of role markers

8

A survey

9

What causes the differences? Plausible factors: • Local discourse traditions • Text length • Sociology of communication • Some structural property of grammar

10

Local traditions • The current sampling does not allow to build this into a statistical model as factor. • But we can analyze a controlled sample of languages that share local traditions (stock of folklore, many conversational strategies, e.g. politeness strategies, rituals, myths, etc.): • Belhare • Maithili • Nepali

• Test for differences between these groups (Bickel 2003) 11

A first hypothesis: local traditions

F (2,27) = 37.0, p = 1.8e-08 (cf. Bickel 2003)

12

Factorial analysis • For the other factors (text length, grammar, sociology), the sample is rich enough for factorial analysis. • But first, what are the factors? What are the predictions?

13

Text length • Perhaps people who talk more use more NPs! • Prediction from Factor “LENGTH”: • LENGTH ~ RD

14

Sociology of communication • Common observation in the Ethnography of Speaking: people who know each other (‘close-knit society’) can presuppose more information than strangers. • This might create a conventional style of ‘presuppositionalistic’ discourse that is also used when talking about unknown referents, as in the Pear Story experiment. • Predictions from Factor “SOC”: • close-knit society ➜ low RD • loose nets, large-scale society ➜ high RD 15

Structural properties of grammar • First guess, although typologically suspect: languages with agreement morphology allow more pro-drop than languages without agreement morphology. • Predictions from factor “M_AGR”: • agreement ➜ low RD • no agreement ➜ high RD

• Another guess (Bickel 2003): structural priming effects from syntax to NP use

16

Structural Priming Effects A construction primes a structurally associated construction (e.g. V → NP; Lu & al 2001). This construction primes its subsequent re-use (e.g. NP → NP; Bock 1986 etc.) Long-term persistence and learning effects (Bock & Griffin 2000) If frequent and strong, this could habituate speakers into a rhetorical norm.

17

Structural Priming Effects (cont’d) •

For this to show, we need a construction that • primes NPs through structural association and that • is frequently used in discourse

•

Such a construction should increase the use of NPs in discourse: Case-based agreement

18

Structural Priming Effects (cont’d) Case-based agreement in Maithili (Indo-European) a. (tũ) 2nhNOM

b. (torā) 2nhDAT

bimār

ch-æ?

sick

be-2nhNOM

khuśi

ch-au?

happy

be-2nhNONNOM

Structural Priming Effects (cont’d) Case-based agreement in Chechen (Nakh-Dagestanian) a. k ant-as

jo

exa

j-o.

b. k ant

jo

exa

j-ie-sh

v-u.

boy[-NOM] girl[-NOM] lie

FEM-do-CVB

MASC-be.NPST

k ant-as

jo -ana

gho

d-o.

boy-ERG

girl-DAT

help

D-do.NPST (= default agreement)

jo -ana

gho

d-ie-sh

v-u.

boy[-NOM] girl-DAT

help

D-do-CVB

MASC-be.NPST

boy-ERG

c.

d. k ant

girl[-NOM] lie

e. k ant-ana jo boy-DAT

FEM-do.NPST

go.

girl[-NOM] see.NPST

20

Structural Priming Effects (cont’d) •

No agreement

•

Case-insensitive agreement in Belhare (Sino-Tibetan) a. (han)

khar-e-ga i?

2sNOM go-PST-2s Q

b. (hanna) un 2sERG 3sNOM

c. ciya

tea.NOM

lur-he-ga

i?

tell-PST-2s

Q

(hannaha) 2sGEN

n-niũa

tis-e-ga

i?

2sPOSS-mind be.easy-PST-2s Q

21

Structural Priming Effects (cont’d) Prediction from Factor “SYN”: Case-based AGR primes argument NPs because they are structurally associated and therefore regularly co-activated; if frequent and strong this should lead to a higher ratio of overt NPs per argument position (= Referential Density or RD): Case-based AGR → high RD NB: the hypothesis is a unidirectional implication, ergo: not case-based agr → low or high RD, depending on other factors, e.g. SOC → Factorial Analysis → Expect interaction between SYN and some other factor(s)

22

Factorial analysis ANCOVA: SOC*SYN*M_AGR*LENGTH Df

MS

F

p

SOC

1

0.09

14.55

0.001

SYN

1

0.05

8.07

0.006

LENGTH

1

0.00

0.00

0.97

M_AGR

1

0.04

6.82

0.011

SOC:SYN

1

0.06

8.97

0.004

SOC:LENGTH

1

0.00

0.59

0.45

SYN:LENGTH

1

0.00

0.44

0.51

LENGTH:M_AGR

1

0.00

0.40

0.53

SOC:SYN:LENGTH

1

0.00

0.00

1.00

51

0.01

Residuals

Adjusted R2 = .339 23

Factorial analysis M_AGR goes against the hypothesized direction…

24

Factorial analysis …and is probably induced by an accidental association of AGR-languages with the highest RD values:

➜ Probably a fake effect; remove from model

25

Factorial Analysis Improved model: SOC*SYN*LENGTH Df

MS

F

p

SOC

1

0.09

14.61

0.00

SYN

1

0.05

8.11

0.01

LENGTH

1

0.00

0.00

0.97

SOC:SYN

1

0.09

14.46

0.00

SOC:LENGTH

1

0.00

0.24

0.63

SYN:LENGTH

1

0.00

0.71

0.40

SOC:SYN:LENGTH

1

0.00

0.09

0.77

53

0.01

Residuals

Adjusted R2 = .342

26

Factorial analysis Interaction effect SOC*SYN:

t=-4.46, df=33.8, p=8.6e-05

ns

27

Factorial analysis Interaction effect SYN*SOC:

t=-5.92, df=15.99, p=2.2e-05

t=-2.07, df=38.83, p=.045

28

Factorial analysis: summary • RD varies systematically depending on both SOC and SYN factors • Each factor is canceled out by the other under high RD levels, i.e. • If SOC yields high RD, SYN has no effect • If SYN yields high RD, SOC has no / a borderline effect • If SOC keeps RD low, SYN has an effect • If SYN keeps RD low, SOC has an effect

29

A closer look at SYN • SYN is based on a priming effect of agreement constructions on structurally associated NPs. • Does the RD effect of SYN directly depend on the actual occurrence of case-based agreement constructions? • We can answer this in Chechen because agreeement varies: • only auxiliaries and ~30% of lexical verbs show agreement (but all agreement is case-based)

30

A closer look at SYN 1. Test correlation between AGR and RD across the ten Chechen narratives.

Coefficent AGR = -.399, p = .12, Adjusted R2 = .16

31

A closer look at SYN 2. Compare RDs in clauses with agreement and those without (paired design): t = -1.53 df = 10 p = .15

32

A closer look at SYN: conclusions • The SYN effect is not directly caused by processing agreement rules. Instead: • Processing agreement rules ➜ priming effects ➜ habituation ➜ conventionalization over time ➜ discourse effects • Independent confirmation: RD trends are amazingly stable in language contact: • Belhare vs. Nepali

33

But how deep are RD differences? • Is it just about pronoun use? Or does it affect more informative NPs as well? • Scale of informativeness: lexical NP > generic NP > pronoun > ø

• Study by Stoll & Bickel (2006) contrasting the most extreme languages in the sample: t = -8.49, df = 18, N = 20, p < .001

34

How deep are RD differences? Comparing RD among lexical NPs only t = 4.49, df = 18, N = 20, p < .001

Corollary: no significant difference in the pronoun/noun ratio. Stoll & Bickel 2006

35

How do they do it? High-RD style in Russian ‘A garden. Near the river a village is visible. The owner is collecting pears in the garden. (He) collected one basket. Another man with a goat appears. The goat is baaing. They went by. The man with the goat went by. And the owner of the garden went to collect the second basket. Here, a boy came towards them on a bicycle, probably his son.’

Stoll & Bickel 2006

36

How do they do it? Low-RD style in Belhare ‘Now, first, (someone) picked apples. (S/he) picked apples and filled (them) into a basket. Then… then, one came along pulling a goat. (S/he) came pulling a goatand took (it) over there across; (s/he) took (it) over there across like that ((gesturing)). (S/he) took away that goat. And then, a bicycle} … one came on a bicycle.’

Stoll & Bickel 2006

37

How do they do it? First mentioning of referents across all 10 Belhare speakers

5

6

17

Stoll & Bickel 2006

38

How do they do it? • In high-RD languages, the names introduced at the beginning serve as reference trackers: ‘Here came a boy towards them on a bike, probably his son. (He) took a basket, put it on the bycycle and carried it away. (He) carried it away. Then, a girl arrived, also on a bycycle. (She) took the second basket. And so. They stumbled on the bicycles. The boy with the basket, with the baskets, with the bags, fell. (Some) guys showed up, who helped the boy to collect the fruits into the basket. (They) lifted it up. And he carried (it) away. So. During the time of collision, when the boy fell, his hat fell off. The guys, who had helped him this one to collect the fruits. went further.’ ’ Stoll & Bickel 2006

39

How do they do it? • Information management in a low-RD language ‘(S/he) came on a bicycle and then (s/he) filled the apples (into the basket) on the bicycle. (S/he) filled in the apples and then went away. (S/he) went away and then again came. Then, another girl came, … on a bicycle. Then, they crashed their bicycles (into each other) and he fell} they fell. They fell and … the girl went over there. As for the boy, he patted his knee right there (where he was). He patted his knee and then four} four (guys) picked up apples for a while, and then one lifted up the bicycle and then (s/he) went away.’

Stoll & Bickel 2006

40

Information management with low RD

• •

All subjects pooled together: Fisher Exact Test, p = .00007 Mean (Lexical) per subject: Wilcoxon Test, p = .004

Stoll & Bickel 2006

41

Information management with low RD • Hold back information as long as possible • Introduce referents only when needed ‘(S/he) came on a bicycle and then (s/he) filled the apples (into the basket) on the bicycle. (S/he) filled in the apples and then went away. (S/he) went away and then again came. Then, another girl came, … on a bicycle. Then, they crashed their bicycles (into each other) and he fell} they fell. They fell and … the girl went over there. As for the boy, he patted his knee right there (where he was). He patted his knee and then four} four (guys) picked up apples for a while, and then one lifted up the bicycle and then (s/he) went away.’

Stoll & Bickel 2006

42

Conclusions •

Typological variance in both RD and RDlex

•

This reflects differences in information management: • free info vs. info only when needed. • cf. collaborative information management reported in the Ethnography of Speaking literature (e.g. Besnier 1989)

•

Perhaps RD differences also have an impact on attentional balance between events and participants.

•

Main causes of the typological variance: • degree of presuppositionalism induced by size of social network • conventionalized style originating from structural priming effects triggered by the way agreement works

•

Experimental demonstration of a ‘deep’ relativity effect from syntax on discourse.

43

Acknowledgements

   

Belhare: Lekhbahādur Rāi Kyirong: Brigitte Huber Nepali: Mādhavprasād Pokharel, Karṇākhar Khativaḍā, Rāmrāj Lohani, Bhimnārayaṇ Regmi Maithili: Yogendra Yādava, Bindeśvar Yādav Chechen: Zarina Molotchieva, Alena Witzlack, Johanna Nichols Xiang: Yan Luo; data from Mary Erbaugh (www.pearstory.org) Russian: Mixail Lurie, Tatjana Krugljakova

 

Swiss National Science Foundation, 1998-2002 University of Leipzig, 2002-2004

  

44